2
$\begingroup$

In this paper Vigeland shows how one can define a group law on subset of a tropical elliptic curve, so that this group is homeomorphic to $S^1$. It is not clear to me what is the relationship between the definition of tropical elliptic curve in the paper to tropicalization of (a toric embedding of a dense open subset of) ellptic curve.

Therefore my question.

Let $A$ be an Abelian variety over a non-Archimedean field $K$, and suppose that value group of $K$ is $\mathbb R$. Is it possible to find an open set $U \subset A$ and an affine embedding $U \subset \mathbb{G}_m^n$ such that the image of $U$ in $\mathbb{R}^n$ under the valuation map can be endowed with a group structure (after restricting to some open subset of this image). If yes, is there a relationship between this group structre and the group structure of the Abelian variety $A$?

$\endgroup$

1 Answer 1

3
$\begingroup$

Abelian varieties have tropicalizations, but the picture is not exactly as you guess.

Let me begin with the case of elliptic curves. So let $A$ be an elliptic curve over a non-archimedean field $K$, let $j$ be its $j$-invariant. Two possibilities arise.

If $|j(A)|\leq 1$, then, possibly after enlarging $K$, it is possible to find a Weierstrass equation for $A$ whose reduction modulo the maximal ideal of the valuation ring of $K$ still defines an elliptic curve. One says that $A$ has good reduction. In some sense, this is the best of all possible cases from the point of view of arithmetic geometry, but unfortunately it escapes the world of tropical geometry.

So assume that $|j(A)|>1$. In this case, whatever Weierstrass equation you choose for $A$, whatever extension of $K$ you take, you won't be able to find one which reduces to the equation of an elliptic curve. However, it will eventually reduce to the equation of a nodal cubic (eg $y^2=x^2(x+1)$). Then, as was observed by Tate, and this is really the beginning of non-archimedean analytic geometry, $A(K)$ can be understood as the quotient of the multiplicative group $K^\times$ by a discrete subgroup of the form $q^{\mathbf Z}$, for some $q\in K$ such that $0<|q|<1$. In his paper « A review of non-Archimedean elliptic functions », Tate has given a wonderful account of this picture in the book Elliptic curves, modular forms and Fermat’s last theorem (J. Coates and S.-T. Yau (eds.) (1995), 162–184).

Tate's point of view is that the classical 19th-century formulas on complex elliptic functions (ie meromorphic functions on $\mathbf C$ with two periods $1$ and $\tau$ — of positive imaginary part), especially Jacobi's, still have some meaning in this framework provided one first mods out by the subgroup generated by $1$ and writes everything in terms of $e^{2\pi i z}$. That is $\mathbf C/(\mathbf Z+\mathbf Z\tau)\simeq \mathbf C^*/q^{\mathbf Z}$, with $q=e^{2\pi i \tau}$.

In rigid analytic geometry, this is interpreted as saying that our elliptic curve $A$ (such that $|j(A)|<1$, recall) has some « $K$-analytic uniformization » by non-archimedean elliptic functions. But now $K^*$ tropicalizes to the real numbers, by the familiar map $z\mapsto \log|z|$, so that $A(K)$ tropicalizes to the real torus $\mathbf R/\mathbf Z\log |q|$. And this map is a morphism of groups.

In Berkovich's point of view on non-archimedean analytic geometry, the analytic space $A^{\rm an}$ associated with $A$ is a (locally contractible) topological space which is simply connected when $A$ has (potentially) good reduction, and whose universal cover is the (analytic space associated with) the multiplicative group when $A$ has multiplicative reduction (ie $|j(A)|>1$).

For abelian varieties of arbitrary dimension, say $g$, the picture is due to Raynaud. For abelian varieties with good reduction, one cannot say much, they are simply connected in Berkovich's sense.

At the opposite spectrum, there are the totally degenerate abelian varieties whose universal cover is a $g$-dimensional torus $\mathbf G_m^g$ for which tropical geometry is a powerful tool. Then, $A$ tropicalizes to a $g$-dimensional torus. This picture has been very important for recent work in diophantine geometry, such as Gubler's proof of the Bogomolov conjecture abelian varieties over function fields which are totally degenerate at some place.

In general, the univeral cover of an abelian variety over a non-archimedean field is an extension of a simply connected abelian variety with good reduction by a torus of some dimension $t$. This gives a kind of partial tropicalization to a real torus of dimension $t$.

$\endgroup$
4
  • $\begingroup$ Thanks for this detailed answer! You have mentioned that as $K^\times$ tropicalises to $\mathbb R$, the quotient tropicalizes to $\mathbb{R}/\mathbb{Z}\mathrm{log}|q|$. What do you mean by "tropicalizes" here? I encountered the following usage of this word: embed into into a torus, then apply valuation map, but it looks like here the map is something else? $\endgroup$ Jun 6, 2014 at 21:06
  • $\begingroup$ Tropicalizing just means applying the valuation map, once it makes sense. The case of semistable formal schemes also gives rise to maps to simplicial complexes associated with the special fiber, but this is a (related but) different story. $\endgroup$
    – ACL
    Jun 6, 2014 at 21:40
  • $\begingroup$ so when one takes $K^\times$ and apply valuation to its factor by $q$, is it possible to somehow relate the quotient to the image of the valuation map composed with some embedding of the quotient elliptic curve (or rather an open subset thereof) into $\mathbb{G}_m^2$, say? $\endgroup$ Jun 6, 2014 at 21:59
  • 1
    $\begingroup$ Yes and no. This question of comparing the skeleton of the elliptic curve (the canonical circle described above) with more naïve tropicalizations from embeddings is discussed in a paper of Baker, Payne and Rabinoff, Nonarchimedean geometry, tropicalization, and metrics on curves, arxiv.org/abs/1104.0320. $\endgroup$
    – ACL
    Jun 7, 2014 at 13:45

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.